[x64] Use vmovapd and vmovsd when AVX is enabled.

src/x64/lithium-codegen-x64.cc

 // Copyright 2013 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #if V8_TARGET_ARCH_X64
 
 #include "src/base/bits.h"
 #include "src/code-factory.h"
 #include "src/code-stubs.h"
 #include "src/hydrogen-osr.h"
@@ skipping 70 matching lines @@
 
 
 void LCodeGen::SaveCallerDoubles() {
   DCHECK(info()->saves_caller_doubles());
   DCHECK(NeedsEagerFrame());
   Comment(";;; Save clobbered callee double registers");
   int count = 0;
   BitVector* doubles = chunk()->allocated_double_registers();
   BitVector::Iterator save_iterator(doubles);
   while (!save_iterator.Done()) {
-    __ movsd(MemOperand(rsp, count * kDoubleSize),
+    __ Movsd(MemOperand(rsp, count * kDoubleSize),
              XMMRegister::from_code(save_iterator.Current()));
     save_iterator.Advance();
     count++;
   }
 }
 
 
 void LCodeGen::RestoreCallerDoubles() {
   DCHECK(info()->saves_caller_doubles());
   DCHECK(NeedsEagerFrame());
   Comment(";;; Restore clobbered callee double registers");
   BitVector* doubles = chunk()->allocated_double_registers();
   BitVector::Iterator save_iterator(doubles);
   int count = 0;
   while (!save_iterator.Done()) {
-    __ movsd(XMMRegister::from_code(save_iterator.Current()),
+    __ Movsd(XMMRegister::from_code(save_iterator.Current()),
              MemOperand(rsp, count * kDoubleSize));
     save_iterator.Advance();
     count++;
   }
 }
 
 
 bool LCodeGen::GeneratePrologue() {
   DCHECK(is_generating());
 
@@ skipping 1866 matching lines @@
     } else {
       // Since we operate on +0 and/or -0, addsd and andsd have the same effect.
       __ addsd(left_reg, right_reg);
     }
     __ jmp(&return_left, Label::kNear);
 
     __ bind(&check_nan_left);
     __ ucomisd(left_reg, left_reg);  // NaN check.
     __ j(parity_even, &return_left, Label::kNear);
     __ bind(&return_right);
-    __ movaps(left_reg, right_reg);
+    __ Movapd(left_reg, right_reg);
 
     __ bind(&return_left);
   }
 }
 
 
 void LCodeGen::DoArithmeticD(LArithmeticD* instr) {
   XMMRegister left = ToDoubleRegister(instr->left());
   XMMRegister right = ToDoubleRegister(instr->right());
   XMMRegister result = ToDoubleRegister(instr->result());
@@ skipping 28 matching lines @@
     case Token::DIV:
       if (CpuFeatures::IsSupported(AVX)) {
         CpuFeatureScope scope(masm(), AVX);
         __ vdivsd(result, left, right);
       } else {
         DCHECK(result.is(left));
         __ divsd(left, right);
       }
       // Don't delete this mov. It may improve performance on some CPUs,
       // when there is a (v)mulsd depending on the result
-      __ movaps(result, result);
+      __ Movapd(result, result);
       break;
     case Token::MOD: {
       XMMRegister xmm_scratch = double_scratch0();
       __ PrepareCallCFunction(2);
-      __ movaps(xmm_scratch, left);
+      __ Movapd(xmm_scratch, left);
       DCHECK(right.is(xmm1));
       __ CallCFunction(
           ExternalReference::mod_two_doubles_operation(isolate()), 2);
-      __ movaps(result, xmm_scratch);
+      __ Movapd(result, xmm_scratch);
       break;
     }
     default:
       UNREACHABLE();
       break;
   }
 }
 
 
 void LCodeGen::DoArithmeticT(LArithmeticT* instr) {
@@ skipping 321 matching lines @@
     __ Cmp(input_reg, factory()->the_hole_value());
     EmitBranch(instr, equal);
     return;
   }
 
   XMMRegister input_reg = ToDoubleRegister(instr->object());
   __ ucomisd(input_reg, input_reg);
   EmitFalseBranch(instr, parity_odd);
 
   __ subp(rsp, Immediate(kDoubleSize));
-  __ movsd(MemOperand(rsp, 0), input_reg);
+  __ Movsd(MemOperand(rsp, 0), input_reg);
   __ addp(rsp, Immediate(kDoubleSize));
 
   int offset = sizeof(kHoleNanUpper32);
   __ cmpl(MemOperand(rsp, -offset), Immediate(kHoleNanUpper32));
   EmitBranch(instr, equal);
 }
 
 
 void LCodeGen::DoCompareMinusZeroAndBranch(LCompareMinusZeroAndBranch* instr) {
   Representation rep = instr->hydrogen()->value()->representation();
@@ skipping 457 matching lines @@
       Register object = ToRegister(instr->object());
       __ Load(result, MemOperand(object, offset), access.representation());
     }
     return;
   }
 
   Register object = ToRegister(instr->object());
   if (instr->hydrogen()->representation().IsDouble()) {
     DCHECK(access.IsInobject());
     XMMRegister result = ToDoubleRegister(instr->result());
-    __ movsd(result, FieldOperand(object, offset));
+    __ Movsd(result, FieldOperand(object, offset));
     return;
   }
 
   Register result = ToRegister(instr->result());
   if (!access.IsInobject()) {
     __ movp(result, FieldOperand(object, JSObject::kPropertiesOffset));
     object = result;
   }
 
   Representation representation = access.representation();
@@ skipping 112 matching lines @@
       key,
       instr->hydrogen()->key()->representation(),
       elements_kind,
       instr->base_offset()));
 
   if (elements_kind == FLOAT32_ELEMENTS) {
     XMMRegister result(ToDoubleRegister(instr->result()));
     __ movss(result, operand);
     __ cvtss2sd(result, result);
   } else if (elements_kind == FLOAT64_ELEMENTS) {
-    __ movsd(ToDoubleRegister(instr->result()), operand);
+    __ Movsd(ToDoubleRegister(instr->result()), operand);
   } else {
     Register result(ToRegister(instr->result()));
     switch (elements_kind) {
       case INT8_ELEMENTS:
         __ movsxbl(result, operand);
         break;
       case UINT8_ELEMENTS:
       case UINT8_CLAMPED_ELEMENTS:
         __ movzxbl(result, operand);
         break;
@@ skipping 50 matching lines @@
     __ cmpl(hole_check_operand, Immediate(kHoleNanUpper32));
     DeoptimizeIf(equal, instr, Deoptimizer::kHole);
   }
 
   Operand double_load_operand = BuildFastArrayOperand(
       instr->elements(),
       key,
       instr->hydrogen()->key()->representation(),
       FAST_DOUBLE_ELEMENTS,
       instr->base_offset());
-  __ movsd(result, double_load_operand);
+  __ Movsd(result, double_load_operand);
 }
 
 
 void LCodeGen::DoLoadKeyedFixedArray(LLoadKeyed* instr) {
   HLoadKeyed* hinstr = instr->hydrogen();
   Register result = ToRegister(instr->result());
   LOperand* key = instr->key();
   bool requires_hole_check = hinstr->RequiresHoleCheck();
   Representation representation = hinstr->representation();
   int offset = instr->base_offset();
@@ skipping 731 matching lines @@
   __ xorps(xmm_scratch, xmm_scratch);
   __ ucomisd(input_reg, xmm_scratch);
   __ j(above, &positive, Label::kNear);
   __ j(not_carry, &zero, Label::kNear);
   __ pcmpeqd(input_reg, input_reg);
   __ jmp(&done, Label::kNear);
   __ bind(&zero);
   ExternalReference ninf =
       ExternalReference::address_of_negative_infinity();
   Operand ninf_operand = masm()->ExternalOperand(ninf);
-  __ movsd(input_reg, ninf_operand);
+  __ Movsd(input_reg, ninf_operand);
   __ jmp(&done, Label::kNear);
   __ bind(&positive);
   __ fldln2();
   __ subp(rsp, Immediate(kDoubleSize));
-  __ movsd(Operand(rsp, 0), input_reg);
+  __ Movsd(Operand(rsp, 0), input_reg);
   __ fld_d(Operand(rsp, 0));
   __ fyl2x();
   __ fstp_d(Operand(rsp, 0));
-  __ movsd(input_reg, Operand(rsp, 0));
+  __ Movsd(input_reg, Operand(rsp, 0));
   __ addp(rsp, Immediate(kDoubleSize));
   __ bind(&done);
 }
 
 
 void LCodeGen::DoMathClz32(LMathClz32* instr) {
   Register input = ToRegister(instr->value());
   Register result = ToRegister(instr->result());
 
   __ Lzcntl(result, input);
@@ skipping 170 matching lines @@
   __ AssertNotSmi(object);
 
   DCHECK(!representation.IsSmi() ||
          !instr->value()->IsConstantOperand() ||
          IsInteger32Constant(LConstantOperand::cast(instr->value())));
   if (!FLAG_unbox_double_fields && representation.IsDouble()) {
     DCHECK(access.IsInobject());
     DCHECK(!hinstr->has_transition());
     DCHECK(!hinstr->NeedsWriteBarrier());
     XMMRegister value = ToDoubleRegister(instr->value());
-    __ movsd(FieldOperand(object, offset), value);
+    __ Movsd(FieldOperand(object, offset), value);
     return;
   }
 
   if (hinstr->has_transition()) {
     Handle<Map> transition = hinstr->transition_map();
     AddDeprecationDependency(transition);
     if (!hinstr->NeedsWriteBarrierForMap()) {
       __ Move(FieldOperand(object, HeapObject::kMapOffset), transition);
     } else {
       Register temp = ToRegister(instr->temp());
@@ skipping 27 matching lines @@
     DCHECK(kSmiTagSize + kSmiShiftSize == 32);
     offset += kPointerSize / 2;
     representation = Representation::Integer32();
   }
 
   Operand operand = FieldOperand(write_register, offset);
 
   if (FLAG_unbox_double_fields && representation.IsDouble()) {
     DCHECK(access.IsInobject());
     XMMRegister value = ToDoubleRegister(instr->value());
-    __ movsd(operand, value);
+    __ Movsd(operand, value);
 
   } else if (instr->value()->IsRegister()) {
     Register value = ToRegister(instr->value());
     __ Store(operand, value, representation);
   } else {
     LConstantOperand* operand_value = LConstantOperand::cast(instr->value());
     if (IsInteger32Constant(operand_value)) {
       DCHECK(!hinstr->NeedsWriteBarrier());
       int32_t value = ToInteger32(operand_value);
       if (representation.IsSmi()) {
@@ skipping 152 matching lines @@
       key,
       instr->hydrogen()->key()->representation(),
       elements_kind,
       instr->base_offset()));
 
   if (elements_kind == FLOAT32_ELEMENTS) {
     XMMRegister value(ToDoubleRegister(instr->value()));
     __ cvtsd2ss(value, value);
     __ movss(operand, value);
   } else if (elements_kind == FLOAT64_ELEMENTS) {
-    __ movsd(operand, ToDoubleRegister(instr->value()));
+    __ Movsd(operand, ToDoubleRegister(instr->value()));
   } else {
     Register value(ToRegister(instr->value()));
     switch (elements_kind) {
       case INT8_ELEMENTS:
       case UINT8_ELEMENTS:
       case UINT8_CLAMPED_ELEMENTS:
         __ movb(operand, value);
         break;
       case INT16_ELEMENTS:
       case UINT16_ELEMENTS:
@@ skipping 37 matching lines @@
     __ subsd(value, xmm_scratch);
   }
 
   Operand double_store_operand = BuildFastArrayOperand(
       instr->elements(),
       key,
       instr->hydrogen()->key()->representation(),
       FAST_DOUBLE_ELEMENTS,
       instr->base_offset());
 
-  __ movsd(double_store_operand, value);
+  __ Movsd(double_store_operand, value);
 }
 
 
 void LCodeGen::DoStoreKeyedFixedArray(LStoreKeyed* instr) {
   HStoreKeyed* hinstr = instr->hydrogen();
   LOperand* key = instr->key();
   int offset = instr->base_offset();
   Representation representation = hinstr->value()->representation();
 
   if (kPointerSize == kInt32Size && !key->IsConstantOperand() &&
@@ skipping 486 matching lines @@
     __ movp(rsi, Operand(rbp, StandardFrameConstants::kContextOffset));
     __ CallRuntimeSaveDoubles(Runtime::kAllocateHeapNumber);
     RecordSafepointWithRegisters(
         instr->pointer_map(), 0, Safepoint::kNoLazyDeopt);
     __ StoreToSafepointRegisterSlot(reg, rax);
   }
 
   // Done. Put the value in temp_xmm into the value of the allocated heap
   // number.
   __ bind(&done);
-  __ movsd(FieldOperand(reg, HeapNumber::kValueOffset), temp_xmm);
+  __ Movsd(FieldOperand(reg, HeapNumber::kValueOffset), temp_xmm);
 }
 
 
 void LCodeGen::DoNumberTagD(LNumberTagD* instr) {
   class DeferredNumberTagD final : public LDeferredCode {
    public:
     DeferredNumberTagD(LCodeGen* codegen, LNumberTagD* instr)
         : LDeferredCode(codegen), instr_(instr) { }
     void Generate() override { codegen()->DoDeferredNumberTagD(instr_); }
     LInstruction* instr() override { return instr_; }
 
    private:
     LNumberTagD* instr_;
   };
 
   XMMRegister input_reg = ToDoubleRegister(instr->value());
   Register reg = ToRegister(instr->result());
   Register tmp = ToRegister(instr->temp());
 
   DeferredNumberTagD* deferred = new(zone()) DeferredNumberTagD(this, instr);
   if (FLAG_inline_new) {
     __ AllocateHeapNumber(reg, tmp, deferred->entry());
   } else {
     __ jmp(deferred->entry());
   }
   __ bind(deferred->exit());
-  __ movsd(FieldOperand(reg, HeapNumber::kValueOffset), input_reg);
+  __ Movsd(FieldOperand(reg, HeapNumber::kValueOffset), input_reg);
 }
 
 
 void LCodeGen::DoDeferredNumberTagD(LNumberTagD* instr) {
   // TODO(3095996): Get rid of this. For now, we need to make the
   // result register contain a valid pointer because it is already
   // contained in the register pointer map.
   Register reg = ToRegister(instr->result());
   __ Move(reg, Smi::FromInt(0));
 
@@ skipping 55 matching lines @@
   if (mode == NUMBER_CANDIDATE_IS_ANY_TAGGED) {
     // Smi check.
     __ JumpIfSmi(input_reg, &load_smi, Label::kNear);
 
     // Heap number map check.
     __ CompareRoot(FieldOperand(input_reg, HeapObject::kMapOffset),
                    Heap::kHeapNumberMapRootIndex);
 
     // On x64 it is safe to load at heap number offset before evaluating the map
     // check, since all heap objects are at least two words long.
-    __ movsd(result_reg, FieldOperand(input_reg, HeapNumber::kValueOffset));
+    __ Movsd(result_reg, FieldOperand(input_reg, HeapNumber::kValueOffset));
 
     if (can_convert_undefined_to_nan) {
       __ j(not_equal, &convert, Label::kNear);
     } else {
       DeoptimizeIf(not_equal, instr, Deoptimizer::kNotAHeapNumber);
     }
 
     if (deoptimize_on_minus_zero) {
       XMMRegister xmm_scratch = double_scratch0();
       __ xorps(xmm_scratch, xmm_scratch);
@@ skipping 58 matching lines @@
     __ CompareRoot(input_reg, Heap::kFalseValueRootIndex);
     DeoptimizeIf(not_equal, instr,
                  Deoptimizer::kNotAHeapNumberUndefinedBoolean);
     __ Set(input_reg, 0);
   } else {
     XMMRegister scratch = ToDoubleRegister(instr->temp());
     DCHECK(!scratch.is(xmm0));
     __ CompareRoot(FieldOperand(input_reg, HeapObject::kMapOffset),
                    Heap::kHeapNumberMapRootIndex);
     DeoptimizeIf(not_equal, instr, Deoptimizer::kNotAHeapNumber);
-    __ movsd(xmm0, FieldOperand(input_reg, HeapNumber::kValueOffset));
+    __ Movsd(xmm0, FieldOperand(input_reg, HeapNumber::kValueOffset));
     __ cvttsd2si(input_reg, xmm0);
     __ Cvtlsi2sd(scratch, input_reg);
     __ ucomisd(xmm0, scratch);
     DeoptimizeIf(not_equal, instr, Deoptimizer::kLostPrecision);
     DeoptimizeIf(parity_even, instr, Deoptimizer::kNaN);
     if (instr->hydrogen()->GetMinusZeroMode() == FAIL_ON_MINUS_ZERO) {
       __ testl(input_reg, input_reg);
       __ j(not_zero, done);
       __ movmskpd(input_reg, xmm0);
       __ andl(input_reg, Immediate(1));
@@ skipping 291 matching lines @@
 
   // Check for undefined. Undefined is converted to zero for clamping
   // conversions.
   __ Cmp(input_reg, factory()->undefined_value());
   DeoptimizeIf(not_equal, instr, Deoptimizer::kNotAHeapNumberUndefined);
   __ xorl(input_reg, input_reg);
   __ jmp(&done, Label::kNear);
 
   // Heap number
   __ bind(&heap_number);
-  __ movsd(xmm_scratch, FieldOperand(input_reg, HeapNumber::kValueOffset));
+  __ Movsd(xmm_scratch, FieldOperand(input_reg, HeapNumber::kValueOffset));
   __ ClampDoubleToUint8(xmm_scratch, temp_xmm_reg, input_reg);
   __ jmp(&done, Label::kNear);
 
   // smi
   __ bind(&is_smi);
   __ SmiToInteger32(input_reg, input_reg);
   __ ClampUint8(input_reg);
 
   __ bind(&done);
 }
@@ skipping 586 matching lines @@
   RecordSafepoint(Safepoint::kNoLazyDeopt);
 }
 
 
 #undef __
 
 }  // namespace internal
 }  // namespace v8
 
 #endif  // V8_TARGET_ARCH_X64